Hook
A $45 billion hedge fund evaporated to $10 billion in months. The same leverage that fueled AI infrastructure speculation is now quietly metastasizing into crypto’s own capital-intensive layer. When the Aschenbrenner fund—backed by a former OpenAI researcher—collapsed under the weight of concentrated AI bets, the market barely blinked. But the forensic trail of that collapse reveals a pattern that should terrify anyone holding illiquid infrastructure tokens. The question is not whether AI spending is slowing; it’s whether crypto’s parallel infrastructure frenzy is already following the same script.
Context
Over the past 18 months, the narrative around AI infrastructure has shifted from “unlimited demand” to “peak capex.” High Capital estimates that by 2026, annual AI-related capital expenditures could exceed $800 billion. Morgan Stanley projects nearly $3 trillion by 2028, with over 80% yet to be deployed. Yet the revenue side remains unproven. The BIS has warned that the spending spree could become a “long-term investment crash.” Meanwhile, the S&P 500’s concentration has reached a half-century high—the top 20 stocks now account for over 50% of its market cap. This concentration is a vulnerability: when capital allocation shifts, the entire index bleeds.
Crypto’s infrastructure story echoes this pattern. The top five Layer 1 blockchains and their associated ecosystems (Ethereum, Solana, Avalanche, Near, and a few others) have absorbed tens of billions in venture funding and token sales. Layer 2 sequencers, data availability layers, and cross-chain bridges have become the new “data centers” of the crypto world. But the revenue generation—transaction fees, MEV extraction, and gas consumption—has not kept pace with the capital deployed. The same dynamic of “defensive capex” (spending to avoid falling behind) is at play, particularly in the race to build modular blockchains.
Core
The mechanics of the AI slowdown are directly transferable to crypto infrastructure. The analysis reveals three key signals that are now flashing in crypto as well:
- Concentration risk: Just as the S&P 500 is dominated by a handful of AI-exposed mega-caps, crypto’s value is concentrated in a few L1s and DeFi protocols. According to CoinGecko, the top 10 tokens by market cap represent over 70% of the total crypto market. This concentration means that a shock to the largest infrastructure projects (e.g., a critical bug in a Layer 2 bridge or a regulatory crackdown on a major sequencer) could cascade faster than in more diversified markets. Tracing the code back to its genesis block—the core assumption that these networks will scale to billions of users—is the same bet that AI investors are making on data centers. If that bet fails, the entire stack collapses.
- Underutilized capacity: The hidden information in the AI analysis is that GPU utilization rates may be falling. In crypto, the equivalent is block space utilization. Many L2s are operating at a fraction of their capacity. Arbitrum, Optimism, and Base combined have a peak throughput that far exceeds current demand. The capital spent on sequencers, sidechains, and DA layers is not yet justified by transaction volume. Where liquidity flows, truth eventually pools—and right now, liquidity is flowing into infrastructure tokens that are priced as if daily active users have already arrived. They haven’t.
- Leverage in the supply chain: The AI analysis highlighted the collapse of the Aschenbrenner fund as a case study of leveraged bets on AI infrastructure. In crypto, we have seen similar dynamics: venture funds borrowing against token holdings to invest in new infrastructure projects, and protocols using treasury tokens as collateral for stablecoin loans. The recent implosion of a $2 billion crypto fund that concentrated on modular blockchains is a microcosm of the same risk. Decoding the signal hidden in the noise—the noise is the excitement around new chains; the signal is the growing debt load on the infrastructure layer.
Contrarian
But here’s the counter-intuitive twist: the AI spending slowdown might actually be a net positive for crypto. The competition for Nvidia’s H100 and B200 GPUs has driven up the cost of hardware for crypto mining and AI services. If AI capex decelerates, GPU supply could loosen, lowering the cost of infrastructure for crypto projects that rely on compute (e.g., decentralized AI networks, zero-knowledge proof generation, and mining). Additionally, capital that was flowing into AI may rotate back into crypto as a more liquid alternative. The Aschenbrenner fund’s collapse was not a failure of the technology but of leverage and concentration. Crypto’s infrastructure layer, while overhyped, is more decentralized and has a track record of surviving bear markets. The blind spot in the AI narrative is that it treats all infrastructure spending as equal, ignoring the fact that crypto’s capital expenditure is often more efficient—it produces a public good (block space) rather than a proprietary data center. Composability is a double-edged sword—but it also means that underutilized capacity can be repurposed for new applications more easily than a physical data center.
Takeaway
Are we approaching the “AI moment” for crypto infrastructure? The parallels are too stark to ignore: concentration, underutilization, and leverage. But crypto’s advantage is that its infrastructure is software-defined and can be recycled. The real question is not whether the bubble will burst, but whether the architecture remains when it does. If the AI slowdown teaches us anything, it’s that the market will punish the most leveraged players first. Watch the smart contracts, ignore the whitepapers. The chain remembers everything.